Chengliang Tang
GU4243/GR5243: Applied Data Science
In digital imaging, the term resolution often refers to the number of pixels in the image.
The convention is to describe pixel resolution with the set of two positive integer numbers, where the first number is the number of pixel columns (width) and the second is the number of pixel rows (height), e.g. 3928 x 6568. Another convention is to cite resolution as the total number of pixels, typically in megapixels, which is the product of pixel columns and pixel rows and dividing by one million, e.g. iPhone Xs Dual 12 MP (megapixel) Camera.
Figure 1. Example of pixel resolution
How to reduce the resolution of an image? Simple - throw away data.
The idea is pretty straightforward: if the original image is too large, we can throw away every other row and every other column (1st, 3rd, 5th, etc) to create a 1/2 size image.
Figure 2. Example of image downsampling
Image downsampling is the most effective way of dimension reduction, and it sometimes helps reduce noise. However, downsampled images inevitably result in information loss.
In many digital imaging applications, due to the limitation of hardwares, high-resolution images are usually unavailable, yet necessary for finer image processing and analysis.
As an application to image restoration, super-resolution (SR) is a class of techniques that construct high-resolution (HR) images from a single or multiple observed low-resolution (LR) images. Recent development in super-resolution techniques have achieved great success in many areas, such as surveillance video, remote sensing, medical imaging and video standard conversion.
According the number of source images for each target image, super-resolution techniques can be mainly divided into two families: single-frame super-resolution and multi-frame super-resolution. In our project, we are focused on single-frame super-resolution.
Figure 3. The effects of super-resolution
Image interpolation, the reverse of downsampling, is the most basic algorithm for super-resolution.
How to increase the resolution of an image? Simple – replicate data. For example, the image is too small, how can we make it 10 times as big? The simplest approach is to repeat each row and column 10 times.